INFORMATION CRITERION-BASED CHANNEL ESTIMATION IN OFDM SYSTEMS WITH UNKNOWN CHANNEL LENGTH

RIVAS HUERTA, KAREN ANDREA (2017)

Catalogado desde la version PDF de la tesis.

Tesis Pregrado

This thesis adresses the problem of channel estimation in OFDM systemswhen the channel length is unknown. This problem includes the joint estimationof the channel and carrier frequency oset (CFO) in the presence ofphase noise (PHN) which correspond to phase distortions in the form of anunknown deterministic variable and a random variable, respectively. Channelnoise variance is also estimated and phase noise bandwidth is assummedknown as well as the transmitted signal.The joint estimation of the channel impulse response (CIR) and the frequencyoset is carried out using Maximum Likelihood estimation. TheExpectation-Maximization (EM) algorithm is implemented due to the presenceof PHN as hidden variable. In the Expectation step, given that PHNhas a nonlinear relation with the output signal, Extended Kalman Filter(EKF) is used as nonlinear lter to calculate the expected posterior distributionof the PHN, whilst the maximization step is carried out by concentratingthe cost in carrier frequency oset, and obtaining the channelestimates in closed form.Akaike's Information Criterion is used as a model selection technique tosolve channel length estimation. The implementation is carried out by usingthree approaches: one direct approach and two others formulated as a regularizedoptimization problem. One of the regularized problems correpondsto the utilization of the `0-(pseudo)norm, whilst the other corresponds tothe utilization of an approximation of the `0-(pseudo)norm.The three approaches are compared considering not only the accuracy ofthe estimation, but the computational load required, in terms of CPU time.EKF was chosen instead of other nonlinear techniques (such as SequentialMonte Carlo techniques) to ensure a fair comparison among dierent AICapproaches.For completeness of the presentation, in this thesis we study the impactof dierent levels of SNR on the overall parameter estimation problem, whenusing full training signals via numerical simulations.